Abstract
Assuming that, in the narrow-band region, the exact Green's function of the Hubbard model essentially retains the two-peak structure proposed by Hubbard, we derive its consequences as rigorously as possible. When a quasielectron is added to the ground state of N electrons in the narrow-band region of the Hubbard model, the whole system is shown to relax due to the strong interaction I, thus yielding a relaxation energy ΔE between the quasielectron and the original N electrons. This explains the conclusion in the preceding paper that a gap ΔE analogous to the energy gap in superconductivity appears between the quasielectron spectrum and the chemical potential for any occupation n regardless of whether the lower band is exactly filled or not and that the system cannot be a normal metal. We further found that the ground state is, in fact, a bound state, and electrons involved do not obey the Landau theory of quasiparticles. However, the electron Green's function describes the motion of an electron added to this N-electron ground state without the relaxation of the original N electrons, and hence this added electron behaves exactly like a Landau quasiparticle and can be represented by the quasiparticle operator Akσ introduced by Luttinger and Nozières. Only if the original N electrons are allowed to relax is the energy of the system reduced by ΔE, yielding the energy gap ΔE. When a quasielectron is removed from the system of N+1 electrons, the same relaxation energy ΔE is removed, and again the gap ΔE is introduced between the chemical potential μ and the quasihole spectrum, thus creating a gap 2ΔE between the quasielectron and quasihole spectra for any occupation n. The Fermi level, lying at the center of the gap, never intersects the quasiparticle spectra and therefore, Luttinger's theorem that the volume within the Fermi surface is independent of interactions and equal to the value in the noninteracting limit does not apply to the Hubbard model in the narrow-band region. However, this conclusion does not immediately imply that the system is an insulator. Instead, it is shown that a current-carrying state analogous to the condensate state found by Fröhlich appears under certain conditions. When the number of electrons N is nearly or exactly equal to the number of atoms Na, quasiparticlelike behavior disappears completely, making the system insulating. The only exception is the pathological limit I=, where a metallic state with a Fermi volume twice that in the noninteracting limit is found.